June 2002

 

Blocks vs. Strips for On-Farm Experimentation

 

Keywords: on-farm trials, strip trials, spatial statistics, replication

 

Jess Lowenberg-DeBoer

 

Introduction

            Farmers want to do their on-farm experimentation in large blocks. At least that is what I have heard in a variety of precision farming focus groups and meetings over the last year. The size of these blocks may be anywhere from a few acres to 40 or more acres, but clearly larger than the strips that have been traditionally recommended for on-farm trials. These strips might be half a planter width, as in the case of “split planter trials,” or one field length pass or one round of a given piece of equipment. This article outlines some of the reasons why producers want to do their comparisons in blocks and some of the issues that blocks raise for decision-making.

 

            Strips are recommended as a way of adapting classical agronomic statistics to the contingencies of on-farm, field level experimentation, where equipment size and time pressure make small plots impossible. Relatively narrow strips have been recommended in an effort to compare treatments on areas that are as similar as possible. The narrow strips are usually essential for replication of treatments within that relatively homogenous area.

           Yield monitoring and other precision farming technologies reduced the cost of on-farm experimentation. Instead of spending scarce harvest time waiting for the weigh wagon and harvesting short ends, the farmer with a yield monitor and GPS could harvest at full speed and spend some time in the winter pondering the yield maps. But on-farm experimentation is still costly. The farmer’s main business is production. Even with precision farming technology, on-farm trials take time and energy away from production. They are justified only if the producer learns something that is more valuable than the time and energy consumed in experimentation. From this cost/ benefit viewpoint farmers are questioning strips.

 

Problems with strip trials

           Among the problems I have heard farmers raise about strip based on-farm trial designs are:

 

·        For hybrid and variety trials, filling planters with small quantities of seed and cleaning boxes for the next hybrid or variety takes too much time.

·        In larger operations, seed is often purchased in bulk. This makes it difficult to fill the planter with small quantities. Hybrid and variety strip trials work better with seed in bags.

·        Split planter trials are convenient only if your combine head is exactly half the width of the planter. That is not always the case.

·        For narrow row soybeans, many producers prefer to harvest at a diagonal to the rows. This makes it impossible to detect narrow strips on the yield maps.

·        In larger farming operations with several family members and/or employees, on-farm trials require coordination. The person planting may not be the person planning. It is easier to communicate the plan when large blocks are treated the same way.

·        Even with GPS guidance, it is difficult to match preplant operations to planter and combine widths. These preplant operations might be tillage, fertilizer spreading or herbicide applications. For example, applying preplant anhydrous ammonia in strips at different rates so that most of the combine passes in the fall are consistently on a given nitrogen rate is difficult for even the most careful operator.

 

Consequences for decision-making

           Comparing treatments laid out in large blocks, rather than strips, has consequences for the reliability of the information and for the decisions based on that information. These consequences occur regardless of how data are analyzed. Most producers that I meet rely on “eyeballing” yield maps for their comparisons. Some calculate averages from whole blocks or polygons within blocks. Very few do statistical tests. But all of them rely on the same principles underlying statistical comparisons:

 

1)      comparing like units, where the treatment is the key source of a difference, and

2)      replication to increase reliability of a conclusion.

 

A key issue is what is the “unit” of comparison? When weigh wagons were used to measure on-farm strip trial yields, the unit of comparison was the yield for the whole strip. With yield monitoring and mapping, we sometimes look at soil types or other management zones in that strip. Visually, we might compare yields of adjacent strips within the same soil type. If the yield of one hybrid or other alternative is always highest then that difference is probably reliable. If in the comparison of the strips it is sometimes one alternative, sometimes the other that has the highest yield, then we say that the difference is not significant. Statistically, classical methods can be adapted to strip trials, though the spatial relationships can cause some problems, especially when soil types or other management zones are included within strips.

For data collected in larger blocks, some alternative units of comparison are:

 

Ž    the entire block,

Ž    soil type or other management zones within the block,

Ž    square cells with side length corresponding to the width of a key piece of equipment (for instance, the planter, fertilizer applicator or sprayer), and

Ž    dots, the raw yield monitor observations.

 

Blocks - If the entire block is used as a unit of comparison, it is likely that the units are quite different from each other. The larger the block the greater the potential differences become. Even in small plot experiments micro-soil variation can wreak havoc with planned comparisons. These differences increase the chances of making a mistake. When comparing yields of hybrids planted in 40 acre blocks, the yield difference could easily be due to the mix of soils, topography or a microclimate created by a windbreak along the side of one block. Blocks as units also severely limit the number of replications and make it harder to separate random “blips” from real differences. Even on large farms it can be difficult to find many similar large blocks.

 

Soil Types - Use of soil types or other management zones can substantially increase the comparability of the units and the number of units observed, thereby potentially increasing the reliability of the comparison. In statistical terms this increases the degrees of freedom. But are these units valid comparisons? In terms of classical statistics they are at best subsamples and at worst “pseudo replication” because they are not independent observations. Yield of a hybrid grown on soil type A will be similar to that grown on neighboring soil type polygon B just because they are close. This similarity may boil down to something as simple as catching the same rain shower at tasseling time.

A comparison of hybrids grown in large blocks on two different soil types may indicate that one of the hybrids produces significantly greater yields. But is this a reliable indicator of yield superiority? Or are they both reacting to some other factor. Is the apparent yield advantage of one hybrid over the other just a seasonal result due to a rain shower that occurred at tasseling time on only one of the blocks.

Another problem is the variability of soils and other factors within any management zone that might be defined. The USDA soil type maps were not designed to be used for site-specific data analysis. The characteristics of a polygon of soil type A in one field is not exactly the same as those of soil type A in another field.  Comparisons of soil type and management zone polygons have many of the same problems as block comparisons.

By taking into account the relationship between neighboring polygons, spatial statistics have the potential for elucidating more information from polygon data than classical statistics, but this approach is still susceptible to problems in defining the zones and the availability of use-friendly software. When Farm Journal asked us to analyze their soil density trial data for the article in the July 2001, issue, my graduate student and I worked part time for over two months to come up with robust comparisons. The soil density data were from soil type polygons from four fields in central Illinois. The unit of comparison was the average yield by soil type. For a magazine that goes to thousands of readers, this may be a good use of time and energy, but with current software no individual producer could justify it. If this approach proves fruitful, software developers could make it much easier to use, just as they have made mapping software easier over the last decade.

 

Cells based on equipment width – Cells would usually be smaller than the management zone polygons, but share many of the same advantages and problems. An example of the cell approach is the analysis of nitrogen trial data by my graduate student, Rodolfo Bongiovanni. In that study the cells were based on the width of the nitrogen applicator, 32 feet. Yield observations within each cell were averaged and assigned to the center of the cell location.

Because they are smaller than zones, cells are potentially more homogeneous within each cell than a management zone would be. Also because cells are smaller there are potentially more units to compare for the same overall test area. As in the management zone case, problems include variability within cells and the correlation between neighboring cells. For the producer or crop consultant eyeballing yield maps, the cells probably create too much information. It would be difficult to make visual comparisons among the 400 one tenth acre cells in a 40 acre block and the 400 cells in a neighboring block.

            Cells have some advantages for spatial statistics. They are of uniform size and shape, as is preferred by certain types of spatial statistics. They can be implemented without the detailed knowledge of the field required for determining management zones. As in the case of management zone polygon analysis, the current software is relatively difficult to use, but could be made more user friendly by a good software developer.

 

Yield Dots – Use of yield dots as units of comparison avoids the problem of defining zones or cells, which is always to some extent arbitrary. The problem is the lack of reliability of those individual dots, which are heavily influenced by combine dynamics and other factors. I am not aware of any statistical analysis which uses the yield dots as the basis for comparison. For visual analysis the number of dots is simply overwhelming. It is impossible to compare the thousands of dots in one 40 acre block with the thousands of blocks in the neighboring 40 acre block. For certain types of spatial statistics, the dots may cause problems because they are not evenly spaced.

 

Questions and some answers:

 

·        Is it better to use blocks or strips for on-farm experimentation? From the standpoint of reliable information, narrow strips are clearly better for most questions. Strips make it easy to do visual comparisons and they lend themselves to the usual statistical analysis.

·        From an economic viewpoint, is it better to use strips or large blocks? That answer is not clear. It is possible that a producer would make more money by focusing on production and using mainly information generated by universities, agribusinesses and his neighbors, complemented by large block comparisons, instead of doing his own strip trials.

·        Is using large blocks for comparisons better than no on-farm information? Most of the time comparisons of large blocks is better than no on-farm information, but the variability within the blocks and the lack of replication may mean that the information is not very reliable.

·        Can technology help us make better use of the data from large blocks? Probably. This is the subject of current research. By taking into account the spatial structure of the data, it may be possible to draw more information out of yield monitor data in those larger blocks.

·        If the methods can be developed to better analyze data from large blocks, can software developers make them easy to use? Yes. Of course, ease of use often comes at a cost. Users do not necessarily understand the analysis and often do not  know whether the defaults are appropriate. But this is no different from the case of mapping software today.      

 

For more information:

 

Bongiovanni, Rodolfo, and J. Lowenberg-DeBoer, “Nitrogen Management in Corn Using Site pecific Crop Response Estimates from a Spatial Regression Model,” Proceedings of the 5th International Precision Agriculture Conference, ASA-CSSA-SSA, Madison, Wisconsin,USA, 2000.

 

Finck, Charlene, Precision Can Pay Its Way,@ Farm Journal, Mid-January, 1998, p. 10-13.

 

Brouder, Sylvie, and Robert Nielsen, “On-Farm Research,” in Precision Farming Profitability, J. Lowenberg-DeBoer and Kathleen Erickson, eds, Agricultural Research Programs, Purdue University, West Lafayette, IN, 2000, p. 103-112.